An entity graph that round-trips data it does not understand, codecs that never import domain semantics, and geometry that lowers into a neutral kernel. No C++ in the dependency graph.
The model stores entities structurally rather than as domain structs. A file full of cost entities parses and re-exports intact in a build compiled with no cost crate at all — verified by tests, not asserted by prose.
How the separations work
A thin viewer takes default-features = false, features = ["step"] and compiles no domain code and no geometry stack. Domains and codecs are cargo features over one shared vocabulary.
Choosing features
Every capability claim names the file that proves it. The capability matrix, the coverage counts and the crate reference are generated from the code, and the build fails when a page drifts from it.
What is covered
ifc-geometry answers what an IFC entity means geometrically and lowers it into the format-neutral Axiolid DAG. Meshing is an opt-in feature with a swappable backend; the default build compiles no kernel at all.
The Axiolid boundary
One core, published to crates.io, npm, PyPI and NuGet, with a versioned C ABI. A file read in one language reads the same in the others, down to the difference between $ and *.
Install
cargo add openbim-ifcThe Install page lists the current release of every package: the Rust crates on crates.io, @openbim/ifc on npm and openbim-ifc on PyPI.
The library target is named ifc, so call sites read as a facade:
use ifc::{Codec, StepCodec};
let model = StepCodec.read_bytes(source)?;
println!("{} entities", model.len());Read the capability matrix before scoping work. It separates what each crate implements from what it deliberately leaves to an application, and each claim cites the file that proves it; its tables are generated from the code. Coverage counts the same thing across the schema.
No capability should be inferred from a crate name, a module path, or an IFC entity appearing in the schema.
If you are evaluating the stack for a specific application, the use cases work an end-to-end scenario against the current state of the code, including what the application author must still build.